OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents
This provides an open resource for researchers to train multimodal models, addressing a bottleneck in the field, though it is incremental as it builds on existing dataset creation methods.
The authors tackled the lack of open datasets for training multimodal models on interleaved image-text documents by creating OBELICS, a web-scale dataset with 141 million web pages, 353 million images, and 115 billion text tokens, and demonstrated its viability by training models that achieve competitive performance on multimodal benchmarks.
Large multimodal models trained on natural documents, which interleave images and text, outperform models trained on image-text pairs on various multimodal benchmarks. However, the datasets used to train these models have not been released, and the collection process has not been fully specified. We introduce the OBELICS dataset, an open web-scale filtered dataset of interleaved image-text documents comprising 141 million web pages extracted from Common Crawl, 353 million associated images, and 115 billion text tokens. We describe the dataset creation process, present comprehensive filtering rules, and provide an analysis of the dataset's content. To show the viability of OBELICS, we train vision and language models of 9 and 80 billion parameters named IDEFICS, and obtain competitive performance on different multimodal benchmarks. We release our dataset, models and code.